Enhancing estimation of water quality index using stacking machine learning techniques: The case of Southern Bug...

Accurate estimation of Water Quality Index (WQI) is essential for sustainable management of water resources. This study explores the use of stacking machine learning (ML) techniques to enhance the accuracy of WQI predictions with considering the Southern Bug River in Ukraine as the case study. The key gap this study attempts to address is the limited application of stacking techniques in WQI predictions, particularly the lack of a comprehensive evaluation of different meta-learners on WQI estimations. While existing studies have explored the use of individual ML models or basic ensemble approaches, this study developed 11 stacking ML methods. The performance of these models was evaluated using a ranking index derived from four statistical metrics. The results indicate that stacking ML models generally outperform standalone ML models. However, it also highlights that the choice of meta-learner significantly influences model performance, as not all stacked models surpass the accuracy of standalone models. To be more specific, the standalone Gaussian Process Regression (GPR) model achieved the best performance in both training and testing phases, obtaining a perfect total ranking index of 1. In addition, reliability and classification assessments demonstrated that standalone GPR attained perfect classification accuracy, while stacking models enhanced the performance of other individual ML models. Nevertheless, computational cost analysis revealed that even though GPR offered the highest predictive accuracy, it also required more computational resources. These findings underscore the potential of stacking approaches for improving WQI predictions, while emphasizing the importance of careful selection of ML models to ensure reliable and efficient water quality management.